Re: [PATCH v4] x86/cpufeature: Add feature dependency checks

[Sohil Mehta <sohil.mehta@intel.com>]

On 2/27/2025 10:46 AM, Ingo Molnar wrote:

>> +void filter_feature_dependencies(struct cpuinfo_x86 *c)
>> +{
>> +	char feature_buf[16], depends_buf[16];
>> +	const struct cpuid_dep *d;
>> +
>> +	for (d = cpuid_deps; d->feature; d++) {
>> +		if (cpu_has(c, d->feature) && !cpu_has(c, d->depends)) {
>> +			pr_info("CPU%d: Disabling feature %s due to missing feature %s\n",
>> +				smp_processor_id(),
>> +				x86_feature_name(d->feature, feature_buf),
>> +				x86_feature_name(d->depends, depends_buf));
>> +			do_clear_cpu_cap(c, d->feature);
>> +		}
>> +	}
> 
> So let's not disable any CPU features actively for the time being, how 
> about issuing a pr_warn() only about the dependency violation?
> 
> I think the main problem is when these problems slip through 100% 
> unnoticed.
> 

I guess you are right. Highlighting the issue is the main part. Beyond
that we can leave the system behavior as-is for now.

Most of the listed dependencies seem to be spec-driven, though the
kernel might create arbitrary dependencies for security reasons such as
making LAM depend on LASS[1]. I think those can probably be handled on a
case by case basis during specific feature enabling.

For the new pr_warn(), I am considering printing it only once per
feature instead of printing it on every CPU (which could be 100s).
But that would mean tracking it in a global feature_warn bitmap.

	DECLARE_BITMAP(feature_warn, MAX_FEATURE_BITS);

Another option would be run the scan only on the BSP. But that could
cause some issues to be missed[2].

I am wondering if there is a better way to do this?

-Sohil

[1]:
https://lore.kernel.org/all/20241028160917.1380714-15-alexander.shishkin@linux.intel.com/

[2]:
https://lore.kernel.org/lkml/ZsfJUT0AWFhoONWf@google.com/#:~:text=divergent%20features%20from%20the%20boot%20CPU